Introduction

The emergence of novel severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was described for the first time in Wuhan, China [1, 2] and declared a public health emergency of international concern on 30 January 2020 [3]. The virus spread rapidly and, as of January 14, 2021, led to 90,759,370 confirmed infections and 1,963,169 deaths worldwide [4].

During the early phase of the pandemic, many countries adopted drastic measures, including testing, tracing, self-isolation, and quarantine measures as well as broader population measures ranging from travel bans, school closures, assembly restrictions, curfews, to full lockdowns [5,6,7]. Besides substantial stressors for individuals and the general public (eg, social isolation, reduced income, restructuring of school, university, and work life) and healthcare systems (eg, disruption of essential health services) [8, 9], the SARS-CoV-2 pandemic has had major socio-economic consequences for the affected countries (eg, global supply chain disruptions) [10, 11]. By drastically changing our way of social interaction (eg, social distancing), it continues to affect many areas of daily life and in line with this social life and participation.

The disease-related threats, containment measures, and associated stressors may have a negative psychological impact on the community at large and potentially even more so on specific risk groups [12,13,14,15,16,17]. Given the work-related stressors in the context of disease outbreaks (eg, high workload, risk of infection, triage decisions), healthcare workers may suffer from a particularly high burden [18]Footnote 1,Footnote 2 [20,21,22]. Patients with pre-existing physical or mental conditions (eg, chronically ill individuals, psychiatric patients, geriatric patients), people with confirmed COVID-19 diagnosis, those recovering from the infection, or suffering from long COVID-19, and subgroups with special risk exposure (eg, caregivers) may also be at risk of developing stress-related mental symptoms [15, 22,23,24,25,26,27,28].

Various systematic reviews have synthesized the evidence on psychiatric symptoms associated with previous highly contagious infectious disease outbreaks (eg, Ebola, SARS-CoV) and the SARS-CoV-2 pandemic [20, 24, 29,30,31,32,33,34,35], some of them also narratively summarizing risk and protective factors for mental health [20, 30, 31, 33, 34]. Several meta-analyses have been conducted, either calculating the pooled prevalence of mental symptoms or odds ratios for the risk of mental burden attributable to the SARS-CoV-2 pandemic [20, 24, 29, 32, 33, 36]. Potential moderators of the negative mental health impact were also partly investigated [32]. International evidence indicates an elevated level of mental symptoms in the general public, including symptoms of anxiety, depression, and stress [30,31,32,33, 36]. Confirming the risk status of healthcare workers, several reviews also found an increased prevalence of mental symptoms in this group [18]1,2 [20, 29, 31, 32]. Finally, a few studies in patient populations (eg, COVID-19 patients, patients with pre-existing mental or physical conditions) show increased mental burden [24, 31,32,33].

There are several shortcomings of reviews published to date. Most either focus on the general population, healthcare workers, or patients, with only few publications examining the level of mental burden across all three specified, most relevant population groups [31,32,33]. Further limitations included a limited search strategy [31], language restrictions [24, 30, 31, 33], or a missing preregistration [20, 24, 29,30,31, 33, 36]. Most importantly, all but one systematic review failed to compare the mental burden during an ongoing pandemic with the burden before the pandemic [31]. Such comparisons, however, are necessary to quantify the mental burden specifically attributable to the current pandemic. We therefore aimed to assess the mental health impact of the SARS-CoV-2 pandemic by comparing data from the early phase of the current pandemic with prepandemic data in the general population, healthcare workers, and patients. We aimed to identify population-specific risk and protective factors for mental health.

Methods

Review registration

This systematic review [37] was preregistered with PROSPERO (registration no. CRD42020193249) with the title ‘Psychological distress, protective factors and resilience during the SARS-CoV-2 pandemic: a systematic review and meta-analysis with comparison to standard data’. Details of the methods are presented in the Additional file 1. The MOOSE Checklist for Meta-analyses of Observational Studies and differences between the protocol and the final review are presented in eTables 1 and 2.

Search strategy and selection criteria

We searched three bibliographic databases from January 1, 2019 to May 29, 2020 (PubMed, PsycINFO, and Web of Science) and inspected the reference lists of included studies. The search strategy comprised terms associated with mental health, pandemics, and the populations of interest (see eMethods 1 in Additional file 1). There were no restrictions concerning language, publication date, or publication format. We did not consider preprint articles. If not reported within a study, we systematically searched for prepandemic comparative data in the same or a similar population (PubMed, PsycINFO; see eMethods 2).

The populations of interest comprised the general population, healthcare workers, and any patients (eg, COVID-19 patients, those with pre-existing physical or mental conditions; eTable 3). Participants were included irrespective of age, health, or employment status. We did not consider infectious disease outbreaks other than due to SARS-CoV-2. To be eligible for the review, studies had to assess at least one mental health outcome, with a broad range of eligible outcomes (ie, anxiety and worrying, depression, posttraumatic stress, sleep, stress, general psychological distress). These outcomes were also considered for a descriptive synthesis of the prevalence (see data analysis). We included original research articles reporting on cross-sectional and longitudinal surveys.

All pandemic studies meeting these criteria were included but were only taken forward to pairwise meta-analyses if using a validated outcome measure and if prepandemic comparative data were available (eTables 4, 5). These were defined as data collected before the exposure to the current pandemic, and in the absence of other disease outbreaks or macro-stressors (eg, disasters), in the same country and population group (if available) and using the same outcome measure. In contrast to the review, we only focused on the four most frequently reported mental health outcomes (primary outcomes), including symptoms related to stress, anxiety, depression, or sleep. Posttraumatic stress, although reported more often than sleep, was not considered for pairwise meta-analyses. As this outcome is usually measured in the aftermath of macro-stressors, we were not able to identify adequate comparative data as mentioned above. Comparative data were selected stepwise using four levels to ensure best available comparability between SARS-CoV-2 exposure (‘pandemic’) studies and prepandemic (‘comparative’) studies. If representative studies in the same country and population (level 1) were not available, we used prepandemic studies in the same (level 2) or an alternative population (level 3; eg, healthcare workers compared with the general population), before resorting to the best available data in a similar country (level 4).

Study selection, data extraction, and quality assessment

The study selection process for the pandemic studies at the level of titles/abstracts and full-texts was performed in duplicate by two reviewers independently (NR, LG). Any disagreements were resolved by discussion or by consulting a third reviewer (KL). At both title/abstract (κ = 0.90) and full-text level (κ = 0.97), excellent inter-rater reliability was achieved.

Relevant information for each included study was extracted in duplicate by two reviewers (NR, LG), working independently, using a customized spreadsheet (eTable 6), which was shortened for the extraction of comparative data. Discrepancies were resolved through discussion or by a third reviewer (KL).

Three independent reviewers (NR, JSW, LG) assessed the quality of included studies using the modified National Institutes of Health (NIH) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies [38] (eTable 7), with disagreements being resolved by discussion or a third reviewer (KL). The level of comparability between pandemic and comparative data was assessed using a self-developed tool with four levels based on the previously mentioned levels for the stepwise selection of comparative data (eTable 8).

Data analysis

The included studies were synthesized in narrative and tabular form, with a descriptive analysis of prevalence rates for mental health symptoms (ie, proportion of participants beyond a cut-off score reported in the included study) and of risk and protective factors. If adequate comparative data for any of the primary outcomes were available, pairwise meta-analyses were performed for the general population, healthcare workers, and patients, respectively (eMethods 3). Given the multiple uses of comparative studies, we used multilevel meta-analyses [39] for the general population and healthcare workers, with pandemic studies being clustered according to prepandemic comparators. For patients, the multilevel model reduces to the classic random-effects model as different comparative studies were available. Prediction intervals were calculated in meta-analyses with at least four studies to take the large between-study heterogeneity into account [40].

Two sensitivity analyses referred to the quality of pandemic studies and the level of comparability (see Search strategy and selection criteria), by limiting the analyses to very comparable pandemic and prepandemic studies (ie, level 1 and 2 mentioned above).

Subgroup analyses for each of the three groups were performed for the surveyed populations (eg, age), characteristics of the pandemic studies (eg, survey start) and of comparative data (eg, publication year), and the relationship of sample sizes in pandemic versus comparative studies, in order to identify potential sources of heterogeneity of the psychological impact of the SARS-CoV-2 pandemic.

Results

Details of the results are presented in the Additional file 2. The systematic search for studies performed during the SARS-CoV-2 pandemic identified 2429 records from database searches and 17 additional records from reference lists, of which 104 studies were included in the review and 43 studies in the meta-analyses (Fig. 1). Of the 104 eligible studies, most studies were performed in the general population (50 studies), followed by 30 studies in healthcare workers, and seven studies in various patient populations. Seventeen studies included mixed samples. Across the three population groups, a total of 208,261 participants ranging from 51 to 52,730 participants [41, 42]1 from the pandemic studies were included in the review, the number of participants considered in the meta-analyses, in total 71,613, ranged from 127 to 60,213 participants (eTable 9).

Fig. 1
figure 1

PRISMA flow diagram

Full size image

The study characteristics of the 104 included pandemic studies (early phase) are presented in Table 1.

Table 1 Study characteristics of included main studies
Full size table

Although we imposed no restrictions on the age limits, we identified no studies conducted in children but did find some studies in the general population that included participants below the age of 18 years [47, 58]1. Thus, the mean age of participants in the pandemic studies ranged from 20 (SD not reported) to 56.9 (SD 7.6) years [49, 99]1. The studies covered Asia (67 studies [26, 41,42,43, 49, 50, 54, 57, 58, 60, 62, 64, 71, 74, 76, 80, 81, 85, 86, 88,89,90,91, 93, 98, 101, 102, 104, 109, 111,112,113,114,115,116, 118,119,120,121,122,123, 126, 124, 128, 131,132,133, 136, 137, 140, 141, 143, 145]1 [18, 48, 59, 73, 84, 87, 92, 100, 105, 106, 108, 117, 134, 138, 139, 142, 144]1,2 [127]1,Footnote 3) thereof from China [42, 49, 50, 57, 58, 60, 62, 64, 71, 80, 81, 85, 86, 88,89,90,91, 93, 101, 104, 109, 115, 116, 118,119,120, 122,123,124, 126, 128, 131,132,133, 136, 140, 141, 143, 145]1 [18, 48, 59, 84, 87, 92, 100, 105, 106, 108, 117, 134, 138, 139, 142, 144]1,2 [127]1,3, Europe (24 studies) [47, 99, 44,45,46, 51,52,53, 55, 61, 63, 65, 66, 68, 70, 110, 77, 78, 82, 83, 94, 97, 130, 103, 135]1, North America (six studies) [56, 67, 79, 107, 146, 147]1, or different continents (seven studies) [69, 72, 75, 95, 96, 125, 129]1. For 13 studies investigating more than one population, several samples were considered [130,131,132,133, 135, 140, 141, 143, 145]1 [134, 138, 139, 144]1,2. We identified 47 matching prepandemic comparative studies (eTable 10), including one pandemic study reporting adequate comparative data [127]1,3 [148,149,150,151,152,153,154,155,156,157,158,159,160,161,162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191,192,193]3.

Prevalence rates of the six mental health symptoms, that were considered for the review, were available for a varying number of included pandemic studies (Table 2). The proportion of participants beyond a cut-off value in the included studies varied considerably (eg, anxiety in general population: 0.7–64.0%). Based on cut-off values reported in the primary studies (eTable 11), we found increased levels of mental burden during the SARS-CoV-2 pandemic in the general population, healthcare workers, and patients regarding each of the symptoms observed during the current pandemic, that is, without considering the prepandemic situation.

Table 2 Narrative synthesis of prevalence based on scores above cut-off values for different mental health outcomes
Full size table

In pairwise meta-analyses comparing pandemic (early phase) with prepandemic data for the four primary outcomes, however, we found only evidence for a small increase of anxiety (standardized mean difference [SMD] 0.40; 95% CI 0.15–0.65; p = .002) and a moderate increase of depressive symptoms (SMD 0.67; 95% CI 0.07–1.27; p = .03) in the general population. No evidence for a change in stress or sleep-related symptoms was identified (Table 3). For healthcare workers compared with healthcare staff before the pandemic, the meta-analyses showed no evidence of any effect on the primary outcomes (Table 3). The same was found for patients (Table 3); however, prepandemic data in patients were only available for four samples. Forest plots are presented in Figs. 2, 3, and eResults 1 in the Additional file 2.

Table 3 Results of main and sensitivity analyses in three populations
Full size table
Fig. 2
figure 2

Forest plot main analysis, general population, anxiety

Full size image
Fig. 3
figure 3

Forest plot main analysis, general population, depression

Full size image

Of the 104 studies, 38 studies were judged to be of fair quality and 57 studies of poor quality, with main concerns regarding selection bias, the validity of outcome measures, and the description of the sample and the survey period (eTable 12). From nine high-quality studies, four were representative surveys [44, 47, 77, 88]1. From the 85 pairwise comparisons relevant for meta-analyses, 52 comparisons were of level-1 and 33 of level-2 quality (eTable 13). When excluding low-quality pandemic studies (Table 3), the effects on anxiety and depressive symptoms in the general population increased. The effect on anxiety in the general population was stable in the sensitivity analysis when only best comparable data sets (ie, level-1 and level-2 comparability) were included, while there was no longer evidence for an effect on depressive symptoms (Table 3 and eResults 2 in Additional file 2).

Heterogeneity was considerable in main and sensitivity analyses, with I2 scores mostly ranging from 90 to 100% and wide prediction intervals (Table 3). We therefore performed subgroup analyses with at least k = 5 studies in the main analyses in attempts to explain this heterogeneity (Table 4; eResults 3 in Additional file 2).

Table 4 Results of subgroup analyses for those populations and outcomes with at least k = 4 studies in main analysis
Full size table

Regarding population characteristics (pandemic studies), age was no consistent risk or protective factor. Within the general population, we identified no evidence for a subgroup difference according to stressor exposure except for elevated sleep symptoms in isolated individuals [62]1. In healthcare workers, there was no evidence for a moderating effect of COVID-19 patient contact on mental health. In different groups of patients, we identified no evidence of differences in anxiety or depression. Compared with COVID-19 patients [131]1, psychiatric patients reported more stress, with the caveat of few studies [42, 132, 135]1.

Among general characteristics of the pandemic studies, we found no (consistent) evidence of differences depending on when the surveys started, whether they were conducted in China, or the sample size. We found evidence of an elevated level of depressive symptoms in the general population and patients depending on the specific outcome measure employed (eg, Patient Health Questionnaire [PHQ], Zung Self-Rating Depression Scale [SDS]).

In subgroup analyses for comparative study characteristics, there was no evidence of a consistent moderation of comparison sample sizes.

Across the three populations, we identified a higher level of anxiety and depressive symptoms if included studies were compared to prepandemic data published five or more years before versus a smaller burden in comparison to prepandemic data of less than 2 years ago.

The relationship of sample sizes explained the heterogeneity of the psychological impact of the SARS-CoV-2 pandemic in the general population and patients, with evidence for elevated symptoms of anxiety if similar sample sizes were compared.

The risk and protective factors narratively identified for each population are presented in Table 5 and eTables 14 and 15, with most of them being investigated in the general population, and few studies investigating protective factors at all. Most frequently named risk factors across the populations were pre-existing mental disorders, female sex, and concerns about COVID-19 infection, whereas most frequently reported protective factors were older age, good economic situation, and higher education.

Table 5 Risk and protective factors in three populations (mostly frequently reported factors)
Full size table

Discussion

To our knowledge, this is the first systematic review and meta-analysis to assess the mental health impact of the SARS-CoV-2 pandemic in the general population, healthcare workers, and patients, by contrasting data from the early phase of the current pandemic with prepandemic data. We identified 104 independent studies, mainly in the general population, that suggest an increased prevalence of mental burden due to the SARS-CoV-2 pandemic. This finding is in line with previous reviews and meta-analyses that merely pooled the prevalence of or calculated the risk for mental burden in either one or several of these groups [20, 24, 29, 32, 33, 36].

On the other hand, the pairwise meta-analyses for 43 studies across the four primary outcomes revealed different results. Compared with prepandemic data, we only found an elevated level of some mental symptoms (anxiety, depression) due to the SARS-CoV-2 pandemic in the general population, but not of stress or sleeping problems.

Although healthcare workers were found to be a group at risk for mental health problems during the SARS-CoV-2 pandemic [18]1,2 [20, 29, 31, 32], we identified no evidence for an increased mental burden during the early phase when comparing them with healthcare staff prior to the pandemic. Because of a (chronic) work-related risk exposure in daily life [194], as a kind of ‘stress inoculation’, healthcare professionals might have learned effective strategies (eg, self-efficacy) helping them to cope more professionally with crises than other groups. In contrast to previous findings [20, 195], the level of COVID-19 patient contact did not affect the mental health impact.

Overall, the results of this review paint a more nuanced picture of the mental health consequences of the SARS-CoV-2 pandemic than previous reviews – an observation in line with stress resilience research that identified different trajectories of psychological adaptation after potentially traumatic events, ranging from no mental burden to severe mental illness [196, 197]. Indeed, a recent analysis of 523 healthy subjects from the German LORA study showed a decrease of perceived stress and stressor load while mental health improved during the eight-week measurement after lockdown, indicating that the pandemic and pandemic response may also have positive effects [198]. The number of studies reporting on protective factors in this review was rather limited, especially in healthcare workers and patients. However, these factors might also partly explain the heterogeneity of findings regarding mental health consequences. This is in line with positive aspects (eg, improved social relationships with close social contacts such as families) that were likewise reported for previous infectious disease outbreaks. The importance of taking a ‘resilience perspective’ in SARS-CoV-2 mental health research and investigating resilience factors has been pointed out previously [19, 22, 197, 199].

Several aspects must be considered when interpreting the results. First, the absence of evidence of effects in healthcare workers and patients in this review does not necessarily mean that there is evidence for the absence of effects of the SARS-CoV-2 pandemic on mental health in these groups. Second, for healthcare workers, the mental burden on individuals probably depends on the location of survey (eg, country, region) and how heavily the respective healthcare systems were burdened in the pandemic timeline (eg, number of hospitalized COVID-19 patients). Among the 13 included studies in meta-analyses for healthcare staff, we could only include a few studies from heavily burdened countries (eg, Italy: k = 2; Spain: k = 0; USA: k = 0). However, nine studies in these meta-analyses had been conducted in China, which, compared internationally, was less affected by the SARS-CoV-2 pandemic [4]. In the subgroup analysis regarding the level of COVID-19 patient contact, we assigned studies to the subgroup ‘high level of contact’ if at least 50% of the sample had close contact to COVID-19 patients (ie, ‘frontline healthcare workers’). However, the nature of contact was insufficiently described in the included studies.

Strengths of this review compared with previous publications include the systematic search for comparative prepandemic data for inclusion in pairwise meta-analyses, the stepwise selection of prepandemic studies to ensure best available comparability, and the population-specific analysis of risk and protective factors. One limitation refers to the search methods for pandemic studies (eg, no preprints; no reference lists of reviews) and comparative data (eg, subgroups in general population only partially searched). We had no restrictions regarding the publication format except for the exclusion of preprints which might be viewed as limitation. This restriction might have affected the evidence found in this review compared to others (eg, Cochrane reviews) where preprint articles are included.

The large between-study heterogeneity, a problem shared by previous meta-analyses [20, 24, 32, 33], could not be fully explained by subgroup analyses. This heterogeneity probably resulted from differences between the pandemic studies (eg, countries, sociocultural differences in the perception of mental burden, pandemic outbreak severity, subpopulations, outcome measures) and variability between the comparative studies (eg, study design, outcome measures), respectively. Among the pandemic studies, especially the specific outcome measures used were an important source of heterogeneity. Furthermore, the pandemic and comparative data were heterogeneous (eg, country, population), which could be partially captured by our self-developed tool for the level of comparability and was controlled for by the corresponding sensitivity analysis. We cannot preclude that moderators of effects are present that we, though our best efforts, did not identify and therefore could not control for. Besides, comparative studies with larger sample sizes were preferred, leading to small 95% CIs and a lack of CI overlap with pandemic study findings. Despite the comprehensiveness of this review compared to previous publications, the small number of studies in certain subgroups potentially limited the statistical power (eg, surveys including students).

Apart from specific outcome measures, less recent comparative data, and homogenous sample sizes, the subgroup analyses indicated no consistent determinants of heterogeneity. An elevated level of depression based on the assessment with the PHQ and SDS might – at least for the PHQ-9 – be explained by the high sensitivity to change of this instrument and its usefulness to monitor treatment outcomes [200, 201]. Given the increased mental burden if pandemic studies were compared to older prepandemic data, cohort effects cannot be excluded.

Discrepancies between subgroup analyses and the narrative synthesis of risk and protective factors (eg, COVID-19 patient contact) might be due to methodological differences. Because of the primary use of screening but not diagnostic tools to determine mental burden in the included pandemic studies, this review does not allow any conclusions concerning a putative increase of diagnoses of mental disorders during the early phase of the SARS-CoV-2 pandemic. Consistent with the synthesis of risk factors, the meta-analyses partly showed an increased level of mental symptoms in young and middle-aged groups, in line with previous studies [12]. However, more studies including elderly would be needed to clearly investigate age differences, and whether the pandemic works as a ‘burning lens’ for the already increased mental burden in young people [202]. Finally, given the pandemic timeline, the evidence is substantially based on Chinese studies thus potentially limiting the transferability of findings to other contexts.

Further research in other countries (eg, USA), that started later on during the pandemic, could change the findings. The latter is also supported by the wide prediction intervals identified in this review, which indicate uncertainty in our conclusions about whether the pandemic and related stressors do affect mental health [203].

The review has several implications for research and practice. There is an urgent need for representative surveys, in order to allow fair comparisons between the mental burden caused by SARS-CoV-2 in different countries and to examine other risk and protective factors (eg, cultural context). Representative surveys in the general population might also serve to identify specific subgroups at risk for which further studies would be needed. From a public mental health perspective, a stronger focus on (psychosocial) protective factors for mental health would be desirable to derive appropriate contents for preventive measures (eg, pandemic preparedness plans) or health-promoting interventions (eg, resilience training) prior to, during, and after a pandemic [199]. By further investigating the mental health impact of specific stressors – in line with Brooks and colleagues [13] – researchers and practitioners might gain further knowledge about when (eg, in pandemic timeline) and for whom (eg, after exposure to which stressors) interventions should be implemented to buffer negative mental health effects of SARS-CoV-2.

Conclusions

In conclusion, compared with prepandemic data, this review shows different adverse mental health consequences of the early phase of the SARS-CoV-2 pandemic in the examined population groups in contrast to previous research, with healthcare workers being more resilient than expected. The quality of studies varies. High-quality, representative surveys in the general population and specific subpopulations, longitudinal studies, and further research efforts on protective factors are needed to better understand the psychological impacts of the SARS-CoV-2 pandemic and to help design effective preventive measures and interventions that are tailored to the needs of specific population groups.